This paper addresses a typical NP-hard problem, the two-dimensional (2D) rectangular packing problem. The study makes improvements on a quasi-human approach, a caving degree algorithm proposed by Huang Wen-Qi, et al., by defining the conception of action space such that the calculation of the caving degree is simplified. Therefore, the evaluation on different placements is reduced considerably, and good layouts could be obtained quickly. The experiments tested 21 famous instances of the 2D rectangular packing problem provided by Hopper and Turton. The improved algorithm achieved optimal layout with a space utilization of 100% for each instance, and the average computing time on a personal computer was within seven minutes. Computational results show that the improvement strategies on the quasi-human caving degree approach are evident and effective.